Semiconductor Manufacturing Trends and Demands

The semiconductor industry’s path to a $1 trillion valuation isn't being blocked by a lack of innovation. Instead, it’s being throttled by a perfect storm of physical and geopolitical constraints.

This represents the most structurally risky period for the industry since the post-COVID-19 correction, as detailed in Omdia’s new SemiDynamics 2026 Q1 Report.

While NVIDIA and Google design ever-more-powerful artificial intelligence accelerators, the reality of 2026 is defined by a brutal shortage of electricity, copper and critical gases. This isn't just a supply chain hiccup; it's a strategic recalibration where scarcity has become the most profitable product.

This constraint is further exacerbated by the "physical AI" inflection, as humanoid robotics and autonomous systems move from pilots to core infrastructure, creating a massive new hardware demand cycle.

The energy chokehold: LNG and the power grid

The five largest hyperscalers — Amazon, Microsoft, Google, Meta and Oracle — have collectively committed more than $660 billion in 2026 capital expenditures, but they’ve hit a wall. AI-optimized data center facilities now require 100-500 megawatts — enough to power entire cities — and the grid simply cannot keep up.

However, the power problem has a new, volatile layer: Liquefied natural gas. In March 2026, disruptions at Qatar’s Ras Laffan hub, caused by missile and drone strikes during the regional war, removed 20% of the global LNG supply.

This has sent electricity costs skyrocketing for energy-intensive fabs in Taiwan and South Korea, and forced a choice between residential heating and chip production.

The U.S. also faces a critical power infrastructure bottleneck: interconnection queues have ballooned to over 2,100 gigawatts — exceeding total grid capacity — while data center developers face reality checks on 2026 timelines. Industry analysis projects 30-50% of planned 2026 data center capacity will slip to 2028. Grid connection processes require three to seven years, and critical equipment like transformers face multiyear lead times.

The material bedrock: Helium, bromine and copper

The industry is grappling with a massive shortage of the raw materials required to actually build and cool these systems.

Helium is indispensable for cooling wafers and leak detection. Following the 2026 strikes on Qatari production —which accounts for one-third of global supply — spot prices have doubled. Fabs in Taiwan and South Korea are now rationing helium. This net supply shortage could lead to a potential reduction in chip production.

Bromine, essential for etching circuits and flame retardancy, has surged to $12,000 per metric ton. With ICL Group in Israel controlling nearly 40% of the global supply, and providing 97% of South Korea’s imports, geopolitical instability has turned this once-common mineral into a strategic liability.

AI infrastructure is incredibly metal-intensive. Each megawatt of data center capacity requires approximately 27 tons of copper for wiring and cooling. Copper prices hit a record $6 per pound in January 2026 and currently sit at approximately $5.61 per pound. Meanwhile, aluminum reached a four-year high of $3,544 per metric ton in early March and remains elevated near $3,505. Data centers are now actively outbidding traditional industrial sectors for these metals.

HBM’s manufactured scarcity and the fab paradox

High-bandwidth memory remains the industry's most lucrative bottleneck.

The market’s biggest manufacturers — SK Hynix, Micron and Samsung, which collectively control production — have preallocated their entire 2026 capacity. Suppliers are reporting record gross margins of 60–70% for HBM, much higher than for standard DRAM. By engineering this scarcity, memory giants have escaped the "commodity trap" that plagued them for decades.

Meanwhile, despite the $400-plus-billion in fab investments from the CHIPS and Science Act of 2022, new capacity is still a dream.

TSMC’s Arizona Fab 2 has been pulled forward to 2027 from 2028, but Intel’s Ohio production starts have been pushed from 2026 to 2030. At the same time, existing fabs are currently offline for "silent rebuilds" as they retrofit lines for high numerical aperture extreme ultraviolet lithography. These upgrades temporarily reduce output, ensuring that supply remains tight enough to justify premium pricing.

The scarcity strategy

The 2026 semiconductor landscape is no longer about who can design the best chip; it’s about who can secure the last tank of helium or the next shipment of copper. This isn't a crisis the industry is rushing to fix.

Scarcity has given semiconductor companies the leverage they’ve always wanted: pricing power. The era of high-volume, low-margin "commodity" chips is over. In its place is a new, strategic model where the most valuable component in an AI chip isn't the silicon — it's the certainty of its delivery.

Investors and manufacturers must prioritize resilience to navigate this "RAMageddon". Adopting just-in-case logistics and design-for-procurement strategies, by securing forward contracts and pin-compatible alternates, is essential to protecting margins and revenue visibility amidst 2026's physical and capital constraints. 

Scientists at the National Institute of Standards and Technology have developed a new way to package photonic integrated circuits so they can withstand extreme temperatures and other difficult conditions.

The advance could allow photonic chip-based technologies to operate in deep-space probes, inside nuclear reactors, in ultrahigh vacuum systems, at near-absolute zero temperatures and in extremely hot industrial settings.

NIST physicist Nikolai Klimov, who led the project, said in an email that some chip manufacturers might adopt this new packaging technique in one to two years in “specialized sectors that urgently need extreme-environment photonics,” such as quantum cryogenic systems, nuclear instruments and hardware used in space.

Demand for advanced semiconductors is rising, prompting the construction of new chip factories and partnerships. Because of increased chip demand, many manufacturers would like use photonic chips in more applications.

However, packaging remains a key obstacle. In chip manufacturing, “packaging” refers to the protective housing and connection system that surrounds a chip and links it to the outside world, including optical fibers, electrical contacts and other components. Good packaging allows the chips to be used in compact, reliable devices without damage or misalignment, NIST said in a news release.

The primary advantage of photonic integrated chips is that they transmit data at high speeds while consuming far less power than conventional chips. Photonic integrated chips currently play a central role in telecommunications, medical diagnostics and advanced sensing.

However, they have had only limited use in more demanding environments because traditional packaging cannot maintain optical connections in the presence of intense radiation, extremely high or low temperatures, or ultrahigh vacuums. Such conditions exist in many places photonic integrated chips could be useful, including quantum computing platforms, space missions, nuclear reactor cores, particle accelerators and industrial and energy applications. 

To allow photonic integrated chips to work in these environments, the NIST researchers needed a new way to reliably attach an optical fiber to a photonic chip. NIST said that standard adhesives like organic polymer glues tend to crack, outgas or degrade when exposed to extreme cold, intense radiation, vacuums or heat. When that happens, the chip can no longer operate.

To solve this problem, NIST scientists adapted a technique called hydroxide catalysis bonding, which NIST said “creates an inorganic, glass-like chemical bond between the optical fiber and the photonic chip.” Instead of relying on glue, the process uses a tiny amount of sodium hydroxide solution to fuse the surfaces at the molecular level, forming a rigid, stable connection.

“Our study marks a major step toward bringing the speed and efficiency of photonics into environments where conventional semiconductor chips powered by electric current and photonics chips packaged using traditional methods have not been able to operate,” Klimov said in the news release.

He added that widespread use in mainstream manufacturing will depend on lowering the price and speeding up production. This could be accomplished in several ways, including reducing the curing time, which currently takes about a week but could be shortened to hours or less; scaling up production; and demonstrating long-term reliability across many devices.

”With these steps, broader commercial adoption could follow in the three to five year timeframe or sooner, depending on how much effort is put into further development of this technology,” he said.

China is outspending the United States in the race to develop cutting-edge chips needed for artificial intelligence and other key technologies, industry representatives said at a hearing of the House Subcommittee on Commerce, Manufacturing and Trade. To compete, the U.S. must build more domestic capacity in all phases of the semiconductor supply chain, they said.

In particular, the representatives called for more “proactive investments” in semiconductor plants and supporting infrastructure, such as reliable sources of electricity. They also said the U.S. should streamline the permitting process for building fabrication plants and otherwise scale back regulations that hamper industry efforts to keep up with China.

“For decades, the U.S. has dominated the world stage of semiconductors — leading in areas such as global revenue, manufacturing capacity and semiconductor design, to name a few," Rep. Gus Bilirakis, R-Fla., chairman of the Subcommittee on Commerce, Manufacturing and Trade, said in his opening statement. “But that leadership is no longer guaranteed.”

Bilirakis noted that American semiconductor manufacturing capacity has declined by more than 25% since 1990. “Our adversaries, particularly China, are actively seeking to unseat us from global leadership and disrupt our supply chains,” he said.

AI race with China not letting up

All of the industry representatives testified that AI is the world’s leading driver of demand for state-of-the-art semiconductors. 

“Leadership in semiconductors will determine leadership in AI,” said Jason Grebe, senior vice president of corporate planning at Intel. “If you do not build the chips, you do not control the future of AI.”

Grebe said Intel has invested more than $100 billion in new and expanded facilities in Arizona, New Mexico, Oregon and Ohio. It also has committed about $93 billion in domestic research and development since 2020. However, the Ohio project has been delayed to 2030, and some have noted that the company continues to have manufacturing capacity issues that will not be easily fixed.

The semiconductor industry is unlike most others because it underpins much of the modern world, said Jason Oxman, president and CEO of the Information Technology Industry Council, which represents information and communications technology companies. He added that the U.S. semiconductor industry “commands over 50% of the global market and generated $318 billion of revenue in 2025, making semiconductors critical to the U.S. economy.”

According to Oxman, semiconductor companies are planning to invest about $1 trillion globally through 2030 in new fabrication plants. Much of that investment will come from China.

“China’s goal is to make its [semiconductor] technology dominant around the world, and its technology stack is increasingly competitive, especially in developing markets,” he said. “U.S. policymakers must enact the right polices to keep the U.S. industry ahead.”

Industrial policy is key

U.S. efforts should focus on establishing a robust industrial policy around semiconductors, said Charles Wessner, a nonresident senior advisor in the Renewing American Innovation Program at the Center for Strategic and International Studies.

“We’ve had industrial policy since about 1787,” he said, pointing to things like agriculture subsidies. “We shouldn’t be afraid to pick winners. [The Chinese] intend to be the winners.”

Wessner added that the U.S. should not try to entirely prevent China or other countries from gaining access to advanced U.S. semiconductor technologies. For example, the MATCH Act closes loopholes that allow countries to obtain restricted technology through front companies, subsidiaries and allied countries. In addition, the Trump administration has placed conditions on Nvidia chip exports to China.

According to Wessner, such strategies can backfire. “They end up being toughest on our own industry” by imposing limits on revenue “that’s absolutely crucial for [U.S. semiconductor companies] to compete,” he said. “This is not an idle point. When you cut revenue by 20%, you cut R&D by 20%.”

While slowing China’s access to certain technologies “may be a useful thing to do, it has to be a targeted effort to slow, not to block,” he added.

“What we need are proactive investments,” Wessner said. “We have to combine our defensive efforts with a careful assessment of their impact. What we have to avoid here is [asking], ‘Who’s toughest on China?’” 

CHIPS and Science Act successes, failures

The 2022 CHIPS and Science Act, which appropriated billions of dollars for semiconductor research and development, has made “visible progress” in boosting U.S. industry, said Asad Ramzanali, director of AI and technology policy at Vanderbilt University.

For example, he said that before that legislation was passed, the U.S. produced no advanced-node logic chips, which include graphics processing units commonly used to train AI models. Almost 90% of those chips were produced in TSMC facilities in Taiwan.

That situation has improved. “By the end of the Biden administration, the U.S. was on track to have 28% of advanced-node capacity by 2032 and had crowded-in over $500 billion in private semiconductor investments,” he told the committee.

However, he said that some U.S. policies under the Trump administration have been counterproductive. These include “gutting CHIPS Act office staffing, conflicts of interest in certain deals and equity investments, including the recent one with Intel, without clear legal authority and transparency.”

As the remainder of CHIPS and Science Act manufacturing funds are distributed, Congress should ensure that they are being used for their intended purpose in the public interest rather than the interest of private entities, he said.

According to the subcommittee chair’s memo, as of March 2026, Commerce has given 46 final awards to 39 different companies, totaling over $29.5 billion in investments. There are a further 14 preliminary memoranda of terms with 12 different companies, totaling over $1.6 billion in potential investments.

Ramzanali also criticized how the Trump Administration has funded research and development through the National Institute of Standards and Technology’s National Semiconductor Technology Center. For example, he said it should not have disbanded key programs like the SMART USA Institute and Industrial Advisory Committee.

The U.S. Department of Commerce has announced it will take a “venture capital-style approach” to NSTC through a broad funding solicitation. However, “That is not what is needed for public investments in CHIPS R&D,” he said.

“Companies already invest vast sums in semiconductor R&D, and this new approach could end up subsidizing what companies would do anyway. Rather, the goal of these funds should continue to be supporting R&D efforts that no one company has an incentive to do alone.”

U.S. chipmakers offered calculated support for the Trump administration’s plan to implement a 25% tariff on certain advanced semiconductors, even as they navigate higher costs while trade negotiations continue.

The tariff, imposed against artificial intelligence chips such as Nvidia’s H200 and AMD’s MI325X, will not affect materials that are imported to support expansion of U.S. semiconductor supply chain and manufacturing capacity, according to the White House. President Donald Trump may impose broader tariffs on semiconductors and products containing them in the near future. The AI chip tariff became effective Jan. 15.

Manufacturing Dive requested comments from a dozen U.S. and international semiconductor companies about the tariff and how they are navigating it. The majority declined or ignored requests for comment.

An Nvidia spokesperson sent an email applauding the president’s decision to support high-paying jobs and manufacturing in the U.S.

“Offering H200 to approved commercial customers, vetted by the Department of Commerce, strikes a thoughtful balance that is great for America,” the Nvidia spokesperson said.

Taiwan Semiconductor Manufacturing Co., the world’s largest chipmaker, and Intel both declined to comment. The U.S. government on Thursday struck a deal with Taiwan, including TSMC, to invest $250 billion in American semiconductor production in exchange for lower tariffs. It also has a 10% stake in Intel valued at $8.9 billion, which President Donald Trump recently touted on Truth Social.

After the trade agreement was made, CFO Wendell Huang told Bloomberg Television that TSMC will continue to grow in the U.S. due to increased customer demand and try to accelerate the transition its semiconductor expertise stateside, but its “leading-edge” technologies and operations will remain in Taiwan.

Other reactions

Global industry association SEMI overall supported the administration’s effort to strengthen U.S. semiconductor production, saying the approach was consistent with the organization’s political advocacy.

Currently, billions of dollars are going toward domestic chipmaking. However, at this point the U.S. is still strongly reliant on foreign suppliers. According to the White House, the nation uses about 25% of the world’s semiconductors and can only make 10% of the chips it requires.

“SEMI urges the Administration to continue engaging closely with industry to ensure any further trade actions account for commercial and operational realities, and minimize unintended impacts on global supply chains,” the organization said in a statement. It noted that a number of international segments cannot be “easily replicated or relocated” quickly enough to support a U.S. buildout.

South Korea, home to major chipmakers Samsung and SK Hynix, is seeking favorable trade terms with the U.S. after Trade Minister Yeo Han-Koo said over the weekend that the new tariff would have a “limited impact” on the country’s companies, Reuters reported.

Many domestic fab projects are underway by the likes of Intel, Micron and TSMC, with many still years away from opening. One of the biggest challenges for chipmakers is that building advanced semiconductor fabrication plants takes a lot of time and capital. Until that manufacturing is moved to the U.S. or the tariffs are reversed, companies that rely on foreign chips for electronics are going to deal with increased prices, said Jack Gold, founder and principal analyst at research firm J. Gold Associates.

“When markets are this unstable, how do you know how to set a strategy for your products?” Gold said. “From a purely manufacturing perspective, there’s a lot of uncertainty in trying to put together a business plan, and that’s not a good thing for companies, and it slows investment, and it causes all kinds of weird stuff going on.”

If someone told Trey Travis five years ago he’d be walking the floor of a supercomputing show, he wouldn’t have believed them.

Georgia-based Southeastern Hose, where Travis is VP of operations, manufactures corrugated metal hoses and traditionally serves customers in industries such as steel and petrochemicals. But the hoses have another purpose: providing cooling and exhaust for data centers. 

As the adoption of artificial intelligence spurred an acceleration in data center construction, orders for his hoses started to trickle in, turning into a deluge last year with another swell of growth since January.

“It's not that the dam opened,” Travis said. “They just took the dam out of the river and it's flowing like crazy.”

Southeastern Hose is among the myriad manufacturers, large and small, seizing on a growing business opportunity with data centers. This comes as other manufacturing industries, such as wood, coal, petroleum, and food and beverage, are contracting.

Conversations with several companies, alongside public reports from many more, show exponential surges in business and double-digit or more increases in revenue as data center construction continues. Some companies are going as far as to establish new business units dedicated to data center customers. 

Spending on data center construction has tripled over the last three years, and occupancy rates remain near record highs for third-party leased data centers, according to Goldman Sachs Research published in August 2025.

Travis said his revenue increased around 25% last year. This year, his business is on track to potentially grow 40%, with about two-thirds of revenue stemming directly from the data center vertical.

While manufacturers are reaping financial benefits, they’re not immune to challenges. Expanding capacity to increase supply and meet demand requires significant capital expenditures, said Bill Pellino, national leader of BDO's manufacturing practice.

“From a cash flow perspective, that can really put pressure on a manufacturer,” he said.

In addition, manufacturers are navigating this amid a tight labor market and rising energy costs, while having to make multiyear business decisions. And no one knows exactly how long the increasingly controversial data center and AI boom will last.

Entering and growing the data center vertical

Darrell West, senior fellow in the Center for Technology Innovation at the Brookings Institution, described AI and data centers as “a boon for American manufacturing,” with the construction boom increasing demand in several sectors. Among the most notable is the semiconductor industry, with chipmakers seeing double-digit or more increases in sales.

For example, Texas Instruments grew Q1 revenue 19% year over year, with data center and industrial demand driving the spike. Intel reported a 22% YOY increase in its data center and AI business unit in the most recent quarter.

In addition, Chipmaker Marvell’s data center revenue grew 21% YOY and 9% over the previous quarter, with the data center market making up about 75% of its business in the latest quarter. Micron’s core data center business unit reached $5.7 billion in revenue, up from $2.4 billion the previous quarter. 

It’s not just semiconductor companies witnessing higher demand.

For Siemens, the data center boom pre-dated AI and started with cloud computing around 2020 and 2021, said Barry Powell, president of Siemens electrical products in North America. The company manufactures switchboards to protect electrical circuits and break down power from utility feeds to individual racks. 

With AI, demand for Siemens products has only ticked up further. Chip manufacturers are producing new generations of chips every six to nine months to support faster processing and longer context windows for AI models. Each chip consumes far more power than the previous generation, Powell said, creating “huge demand” for energy and electrification products.

In the company’s most recent quarter, orders rose to a record high, mainly due to the electrical products business and larger contracts with data center customers in the U.S. To keep pace, Siemens and other manufacturers are in “a race to ramp up capacity,” Powell said.

Hyperscalers want fast times to market and have standardized their designs to make that happen. In turn, Siemens standardized its manufacturing processes, building dedicated factories to take on high-volume, low-mix electrical equipment. Meanwhile, other factories serve Siemens’ non-data-center verticals, with lower volume and higher mix.

Many of Siemens’ customers are locking in two- to three-year contracts with commitments to dedicated volume over multiple years. 

“That allows us to go make these investments that are incredibly expensive with a shared amount of risk,” Powell said. “This is not a fly-by-night thing we think is going to slow down in the next year or two.”

In March, Siemens invested $165 million to expand its manufacturing facilities in North and South Carolina, and it opened a 500,000-square-foot, $190 million site in Fort Worth, Texas, to meet data center demand. Siemens also launched a training program with the latter facility. The company brought in workers from jobs like retail or food delivery and trained them on manufacturing skills. 

Multimillion-dollar investments are part of ABB’s plans as well. The electrical manufacturer announced last year it would invest an additional $110 million in the U.S., including a new production line in Mississippi for data center circuit breakers and increased capacity in North Carolina and Virginia to support data centers and other customers.

ABB provides electrical infrastructure that sits within or right outside of data centers, such as distribution equipment, switch gears, sensors and circuit breakers. Data centers have been a segment for ABB for more than a decade. Within the last 12 months, orders in the vertical have grown more than 200% – the highest of any segment, said Amanda Trumble, data center segment leader of the Americas.

Recently, electrical component manufacturer Eaton created a new business segment specifically focused on the data center market. Eaton invested $50 million last year into a new Virginia facility for power distribution technologies to help meet data center demand. Virginia is home to 35% of hyperscale data centers globally.

Likewise, steelmaker Nucor established a dedicated business unit called Nucor Data Systems to manage orders coming in from hyperscalers and developers. And Cleveland-Cliffs is exploring the sale of idle mills to potential buyers, which could include data center developers.

In another example of the many sectors involved, Corning in January signed a $6 billion agreement with Meta Platforms to expand manufacturing capabilities across its North Carolina facilities and supply Meta with optical fiber and cables.

Thermal management and cooling are also critical for data centers, due to the heat chips generate.

As AI and GPUs become more advanced and energy intensive, “they run super hot,” said Ross Toepel, national project development manager at Krohne, which manufactures flowmeters. “Air cannot cool them fast enough.”

This means they need liquid cooling. Krohne’s devices measure the flow of water to the chip; cool water flows in, while warm water from the chip’s generated heat flows out and recirculates.

Krohne has historically served industries such as food and beverage, chemical, and oil and gas. But in March 2025, Toepel started looking into what opportunities might exist for the manufacturer in the data center vertical. His conclusion: “This is the biggest opportunity that our company will see … probably in the history of the rest of our lifetimes.”

Today, Toepel estimates the data center vertical makes up around 20% of its business.

“We're growing our business in a way that we haven't seen before, because we've never had access to this industry,” he said.

Some manufacturers are also developing products for data centers that combine multiple elements. Vertiv makes a “OneCore” product that combines power modules and chillers into one unit.

“Customers need partners who can deliver power, thermal management, and services as one integrated system,” Anand Sanghi, president of the Americas for Vertiv, said in an email.

In March, Vertiv announced a $50 million expansion investment in Ohio to increase production capacity by about 45% for liquid cooling and chilled water systems. The manufacturer reported Q1 net sales up 30% YOY with “strong data center demand” contributing to revenue growth. 

BDO’s Pellino said many manufacturers are using AI to predict business. This creates a loop in which data centers fuel AI for business planning, which is then used to manufacture products for data centers.

The quick pace has also led to “some very strategic M&A to get the capabilities in-house that you may not otherwise have,” Pellino said.

All in the last year, Eaton acquired Boyd’s thermal business for $9.5 billion. The purchase enabled Eaton, which previously only offered air cooling, to enter the liquid cooling market.

Samsung Electronics inked a deal to acquire Germany-based FläktGroup for about $1.7 billion. The latter is a heating and cooling systems provider, bolstering Samsung’s ventilation and air conditioning business for data centers. Nucor acquired Southwest Data Products, a manufacturer and installer of data center infrastructure, for $115 million.

Long-term growth, near-term challenges

With the clear promise of growth and financial benefits, businesses that aren’t yet supplying data centers are examining how they can get into the vertical, Pellino said. For example, a company that’s already making chillers or heat exchangers can focus on supplying those components to data centers.

Pellino advised manufacturers to treat data centers like any other vertical:  Assess the life cycle of the sector, what percentage of the total business should occupy the vertical, the opportunity cost of the decision and how innovation could change the relevance of their product. For example, if a company manufactures cooling towers, the advancement of self-cooling chips could render that manufacturer obsolete for data centers.

“They need to go in with the strategy on how this fits into their business rather than looking at it as a short-term opportunity to gain revenue,” he said.

Southeastern Hose projects sustained demand for the next four to five years. It recently completed a 40,000-square-foot expansion on a manufacturing facility, which was initially intended for the build-up it saw during the pandemic through 2024. Now the manufacturer is looking at additional facilities.

The challenge with rapid capacity expansions, is finding sufficient talent to staff the plant. Travis said he’s doubled his workforce over the last four years, adding hires from the leadership team to the shop floor. The manufacturer has been around for 40 years “so we’re well known” in the local community, he said. The company hires through word of mouth and partnerships with nearby schools.

Future Form, a Nevada-based manufacturer of doors, frames, cabinets, server racks and other components for data centers, hasn’t had to significantly scale up its workforce because it has invested in automation and smart manufacturing.

The manufacturer entered the data center vertical a decade ago, but orders were small overall, said Brian Delnevo, director of supply chain. Over the last five years, “we've seen an astronomical uptick in volume of orders,” he said. “It’s been going gangbusters.”

The company tripled the square footage of its warehouse and manufacturing space over the last two years. It also added a 3D printer to produce smaller parts and pieces primarily for the data center vertical. Delnevo said the printer, which came online in April, will allow Future Form to produce products at a much faster rate and “drastically reduce pricing for customers.”

Delnevo said Future Form works directly with mills to source cold-rolled steel. It secures the amount of material needed one to two years out. It mostly sources its steel domestically and tries to “steer clear of importing as much as possible,” Delnevo said.

Tariffs on metals have raised the price of data center construction, West of Brookings said. Any materials passing through the Strait of Hormuz could create delays or disruptions down the supply chain.

Powell of Siemens noted higher prices in key elements. For example, copper prices are at record highs, according to the International Energy Agency. Prices for silver, which goes in contacts for electrical equipment, are averaging double what they were last year, per J.P. Morgan. 

In addition, “the lead time for this whole industry has gone up tremendously,” Powell said, adding that low-voltage equipment has lead times around nine months, and medium-voltage is around one year. 

The firms that spoke to Manufacturing Dive in recent weeks said they haven’t faced supply constraints for their materials, but chip manufacturers such as Micron and Broadcom have noted supply constraints creating a gap between supply and demand.

A report from Omdia said high-bandwidth memory used in data centers is sold out through the end of the year. Producers are prioritizing AI infrastructure over consumer electronics or industrial needs because margins for high-bandwidth memory are much higher – around 60% – compared to commodity DRAM, which is around 20%. For manufacturers, “there is no economic incentive to reverse course” the report stated, which leaves PC and smartphone manufacturers struggling for supply.

Pressure on manufacturing and the unknown future

At the same time, the data center boom could also complicate situations for manufacturers that aren’t involved in the vertical. 

Any time demand begins to outpace supply for components, “manufacturers should expect to see their costs rising,” Pellino said.

Plus, data centers “are sucking up all the electrician talent,” West said, noting that a manufacturing employer in need of electricians for infrastructure development, construction or building projects will have “a hard time finding” the talent. The Bureau of Labor Statistics predicts 81,000 openings for electricians each year through 2034.

With data centers consuming immense power and significant water resources, industrial electricity rates are up year over year in nearly every region of the U.S. except the West Coast, Alaska and Hawai’i.

Strain on the grid could worsen as Goldman Sachs Research projects global power demand from data centers will rise 50% by next year and 165% by 2030. West said local water rates have increased in some areas, impacting both residents and other manufacturers pulling the same streams of water supply, with shortages possible in drought-prone areas. 

Some communities and lawmakers have also started to push back against hyperscale projects and data center construction. If resistance grows, data center construction could slow with ripple effects to the manufacturers supporting that vertical. 

Additionally, manufacturers that focus too much on the data center market may run the risk of alienating their long-term customer base. 

“They don't want to screw over all of their existing customers because of the new data center demands. That would probably be a risky business proposition,” West said.

Travis said Southeastern Hose is doing its best to still accommodate industries like steel, oil, petrochemicals and HVAC. But data center customers are “willing to pay whatever they need to pay” to be first in line for components, he said. The manufacturer has had conversations with long-time customers explaining the changing market and greater workload for the metal hose industry.

No one truly knows how long data center demand may last. Siemens anticipates the fast pace of growth continuing through 2028 or 2029, at which point the market will still grow but at a slower pace. Powell estimated a rate of 6-8% growth, versus the 15-20% growth he’s observing today. 

A recession or continued geopolitical conflict could also push data center construction off course. And If the AI bubble were to burst sooner than anticipated, it would create “a huge problem for all the manufacturers who were assuming a certain level of demand over a number of years,” West said.

Toepel said if demand does dry up, Krohne can quickly pivot to the wastewater industry, which uses the same technology. Travis said his product is consumable and will eventually need to be replaced in existing data centers, giving him some guarantee of business. He predicts a plateau around 2030.

ABB’s Trumble sees a long-term strategy from data centers. Even if AI growth stalls or slows, she said, data centers still serve as the backbone for the internet and the cloud.

“There's no way that that requirement is going away.”

Intel’s first-quarter revenue was up 7% year over year, while gross margin was 39.4%, an increase of 2.5 percentage points from Q1 2025.

Artificial intelligence is “moving into the real world” toward applications like “agentic, physical AI, and robots and edge AI,” CEO Lip-Bu Tan said on an April 23 earnings call. He said the shift has started to drive strong demand for central processing units, including the company’s x86 ecosystem.

Intel’s AI-driven businesses grew 40% year over year and now represent 60% of revenue, CFO Dave Zinsner said.

Intel’s Q1 results “demonstrate continued and steady progress across the business, reflecting strong demand for our products and disciplined execution to expand available supply,” Tan said. He added that revenue, gross margin and earnings per share were all above the high end of guidance, and that the company has exceeded financial expectations for the sixth consecutive quarter.

According to Tan, demand continued to exceed supply in Q1 for all of Intel’s businesses, especially for Xeon server CPUs. A CPU-anchored architecture remains the “backbone of AI computing in production,” he said, which is “a structural reason I am confident that CPU franchise will continue to be a meaningful growth engine for the company in the years ahead – not just the quarters ahead.”

“Intel is now a very different company than when I first joined over a year ago,” Tan said. “We have taken, and continue to take, deliberate steps to rebuild Intel into a more competitive and more profitable company.”

The Client Computing Group unit earned $7.7 billion in revenue in Q1, down 6% sequentially but up 1% year over year. Data Center and AI earned $5.1 billion, up 7% sequentially and 22% year over year.

Tan highlighted several developments in Q1, including progress with Intel Foundry, which expanded advanced packaging and testing capacity at a facility in Penang, Malaysia. Although the foundry’s initial results are encouraging, realizing its full potential will be a “long journey,” he said.

Intel Foundry earned $5.4 billion in Q1 2026, an increase of 20% sequentially and 16% year over year. The foundry had an operating loss of $2.4 billion; this was a slight improvement of $72 million quarter over quarter, Zinsner said.

Tan also cited a multiyear collaboration with Google on AI infrastructure. According to a press release, the agreement calls for continued use of Intel’s Xeon processors in Google Cloud infrastructure across AI, inference and general-purpose workloads. The deal “reinforces the central role of CPUs and [infrastructure processing units] in modern, heterogeneous AI systems,” the release said.

In addition, Tan pointed to Intel’s partnership with Elon Musk-owned SpaceX, xAI and Tesla, announced on X. Under the deal, Intel will join Musk’s Terafab project to help build custom-designed chips. 

“Elon and I share a strong conviction that global semiconductor supply is not keeping pace with the rapid acceleration in demand,” he said. “We are excited to explore innovative ways to ‘refactor’ silicon process technology – looking for unconventional ways to improve manufacturing efficiency that will eventually lead to a dynamic improvement in the economics of semiconductor manufacturing.”

Looking ahead, the company expects second-quarter revenue between $13.8 billion and $14.8 billion.

“Q1 was a strong quarter financially and operationally,” Zinsner said. “All demand signals continue to emphasize the growing and essential role of the CPU in the AI era and the unprecedented demand for leading edge wafers and advanced packaging.”